If Artificial Intelligence often makes mistakes – we all know that – human intelligence also has its embarrassing limits (unless one wants to look for something more complex and less conscious, of a Freudian nature) and so it has happened that along with version 2.1.88 of the Claude software, Anthropic’s flagship artificial intelligence, the company also distributed the source map, nearly 2,000 TypeScript files and more than half a million lines of code: a real delicacy that tech enthusiasts (but less skilled ones can use other AIs for this purpose) could use to reconstruct the entire Claude Code source code.
Claude code source code has been leaked via a map file in their npm registry!
Code: https://t.co/jBiMoOzt8G pic.twitter.com/rYo5hbvEj8
— Chaofan Shou (@Fried_rice) March 31, 2026
AN UNINTENTIONAL LEAK
For a few hours there was even talk of a hacker attack, since for the company it is a major own goal that will allow everyone, both rivals and potential malicious actors, to sift through the folds of an intelligent algorithm that is raising billions upon billions in funding rounds. A few hours after the buzz on the Net, it was the Amodei brothers’ software house itself, the true rival of OpenAi’s ChatGpt after all the commotion with the Pentagon, that intervened and outlined the perimeter of the incident.
“This release – reads the official statement addressed to VentureBeat.com – was caused by human error, it is not a security breach. We are implementing several countermeasures – it is explained – to prevent it from happening again.” No malicious attack then, but a sensational and perhaps even more embarrassing human oversight.
TOO MANY INCIDENTS AROUND CLOUDE?
However, this is not the first incident of this kind to involve the Amodei brothers’ company (in the photo Dario): just a few days ago, in fact, the uncontrolled dissemination of private documents allowed the world to discover in advance that Anthropic is working on a new Artificial Intelligence model internally called Mythos.
It would be a particularly powerful product and so far the leak could constitute a mere over-the-top marketing operation, however the leak also revealed that this new algorithm still has significant security issues. And this certainly does not advertise Anthropic’s new AI in the best light, which however at least for now seems to have other problems to deal with: the leaks, both involuntary and not.
